An Empirical Study on Sentiments in Twitter Communities

Abstract

Sentiment analysis and community detection are two popular research subjects in data mining. Lots of research have been published in recent years that aim to enhance the mining of text using sentiment analysis tools and to mine network structure to find cohesive and important communities in social networks. However, there is a lack of knowledge of the importance of understanding the sentiment and its changes on the community lifetime. In this paper, we aim to study the sentiments and its impact on user behaviour and the evolution of social network communities. To do that, we collect three Twitter datasets, two of which are based on the communications between people who share following links and the third dataset is based on people who talked about world cup subject. Next, we analyse the sentiments of communications to positive, negative or neutral. After that, we detect communities using k-core. Later, we track changes of sentiments in communities for an extended period of time. Our results showed that the positive sentiment is contagious because members of the communities increasingly share positive tweets more than the negative ones over time. Also, we found a strong correlation between positive sentiments and the size of the community in all three datasets. These results lay shed on the existence of like-minded users within the communities which attract social network companies for their viral marketing and recommendation systems.